Phylogenetic Relationships of Ruteae (Rutaceae): New Evidence from the Chloroplast Genome and Comparisons with Non-Molecular Data

Home , Boenninghausenia, Choisya, Choisya ternata, Cneoridium, Dictamnus, Haplophyllum (plant)

ARTICLE IN PRESS

Molecular Phylogenetics and Evolution xxx (2008) xxx–xxx

Contents lists available at ScienceDirect

Molecular Phylogenetics and Evolution

journal homepage: www.elsevier.com/locate/ympev

Phylogenetic relationships of Ruteae (Rutaceae): New evidence from the chloroplast genome and comparisons with non-molecular data

Gabriele Salvo a,*, Gianluigi Bacchetta b, Farrokh Ghahremaninejad c, Elena Conti a a Institute of Systematic Botany, University of Zürich, Zollikerstrasse 107, CH-8008 Zürich, Switzerland b Center for Conservation of Biodiversity (CCB), Department of Botany, University of Cagliari, Viale S. Ignazio da Laconi 13, 09123 Cagliari, Italy c Department of Biology, Tarbiat Moallem University, 49 Dr. Mofatteh Avenue, 15614 Tehran, Iran article info abstract

Article history: Phylogenetic analyses of three cpDNA markers (matK, rpl16, and trnL–trnF) were performed to evaluate Received 12 December 2007 previous treatments of Ruteae based on morphology and phytochemistry that contradicted each other, Revised 14 July 2008 especially regarding the taxonomic status of Haplophyllum and Dictamnus. Trees derived from morpho- Accepted 9 September 2008 logical, phytochemical, and molecular datasets of Ruteae were then compared to look for possible pat- Available online xxxx terns of agreement among them. Furthermore, non-molecular characters were mapped on the molecular phylogeny to identify uniquely derived states and patterns of homoplasy in the morphological Keywords: and phytochemical datasets. The phylogenetic analyses determined that Haplophyllum and Ruta form Ruta reciprocally exclusive monophyletic groups and that Dictamnus is not closely related to the other genera Citrus family Morphology of Ruteae. The different types of datasets were partly incongruent with each other. The discordant phy- Phytochemistry logenetic patterns between the phytochemical and molecular trees might be best explained in terms of Congruence convergence in secondary chemical compounds. Finally, only a few non-molecular synapomorphies pro- Shimodaira–Hasegawa test vided support for the clades of the molecular tree, while most of the morphological characters tradition- Character mapping ally used for taxonomic purposes were found to be homoplasious. Within the context of the phylogenetic Homoplasy relationships supported by molecular data, Ruta, the type genus for the family, can only be diagnosed by using a combination of plesiomorphic, homoplasious, and autapomorphic morphological character states. Ó 2008 Elsevier Inc. All rights reserved.

1. Introduction More generally, the choice of characters for phylogenetic analysis has been a crucial and controversial issue in systematics (e.g., Testing whether traditional taxonomic classifications based on Hart et al., 2004; Stace, 2005) and the relative role of molecular morphology are congruent with more recent molecular phyloge- and morphological data in reconstructing phylogenies has been netic findings has become a central task in the current systematic extensively debated (Hillis, 1987; Patterson, 1988; Sytsma, 1990; agenda (e.g., Simões et al., 2004; Van der Niet et al., 2005; Wiens Donoghue and Sanderson, 1992; Novacek, 1994; Baker et al., et al., 2005; Marazzi et al., 2006; Rønsted et al., 2007; but see 1998; Wahlberg and Nylin, 2003; Wortley and Scotland, 2006). Di- Grant, 2003). Disagreements between morphological taxonomies rectly linked to character choice is the controversy about combined and molecular phylogenies have often been attributed to high lev- versus separate analyses of different datasets (Bull et al., 1993; de els of homoplasy in characters traditionally used to delimit taxa Queiroz et al., 1995). For example, should morphological, molecu- (e.g., Lavin et al., 2001; Moylan et al., 2004; Mueller et al., 2004; Si- lar, and phytochemical characters for a certain group of organisms mões et al., 2006) and taxon diagnoses based on plesiomorphic be analyzed together or separately? Advocates of separate analyses morphological character-states (e.g., Roalson et al., 2005; Norup have stressed the fact that congruence among trees derived from et al., 2006). Incongruence between molecular phylogenies and independent sources of data can offer strong evidence for the accu- morphological classifications has prompted the recognition of racy of the inferred relationships (Swofford, 1991; Hillis, 1995; groups highly supported by molecular data, but lacking unique Miyamoto and Fitch, 1995; Graham et al., 1998), while incongru- morphological synapomorphies (e.g., Porter and Johnson, 2000; ence can provide initial insights on important biological phenom- Lavin et al., 2001; Hughes et al., 2004), or the dismantling of tradi- ena, ranging from hybridization to lineage sorting (e.g., Rieseberg tionally accepted taxa (e.g., Kim et al., 1996; Kron et al., 1999; et al., 1996; Won and Renner, 2003; Doyle et al., 2004). Conversely, Wiens et al., 2005). advocates of global evidence have emphasized the fact that combining datasets before phylogenetic analysis grants the best oppor- * Corresponding author. Fax: +41 446348403. tunity to resolve relationships at different scales of divergence E-mail address: [email protected] (G. Salvo). (Cunningham, 1997; Kluge, 1998; Gatesy and Baker, 2005).

Please cite this article in press as: Salvo, G. et al., Phylogenetic relationships of Ruteae (Rutaceae): New evidence from ..., Mol. Phylogenet. Evol. (2008), doi:10.1016/j.ympev.2008.09.004 ARTICLE IN PRESS

2 G. Salvo et al. / Molecular Phylogenetics and Evolution xxx (2008) xxx–xxx

Ruta L. (Rutaceae Juss.) and related genera offer a primary Ruta (around 60 species) by tetra- and pentamerous flowers, a example of the discordant systematic conclusions that can be thick cushion-shaped nectary disk, and dorsally angled seeds; reached by using different types of data. Below we provide the nec- Thamnosma (one species) by almost reniform seeds and variation essary background to understand the sources of such discrepancy in the shape of the nectary disk; Boenninghausenia (one species) and explain how novel evidence from molecular characters might by a cup-shaped nectary disk and filiform filaments; Cneoridium help to clarify the discordant taxonomic treatments published un- (one species) by one carpel, two ovules per locule, and an almost til now. The paucity of diagnostic morphological traits, combined spherical stigma; Psilopeganum (one species) by a relatively small with their overlapping and contradicting nature, has hindered both nectary disk with a narrow ending; and Dictamnus (one species) a stable circumscription for Ruta—alternately subjected to taxo- by zygomorphic flowers, lanceolate petals and sepals, club-shaped nomic ‘‘lumping” (Engler, 1896, 1931) and ‘‘splitting” (Townsend, filaments with protruding glands, and three ovules per locule (see 1968, 1986)—and the unequivocal identification of relationships Table 2). Psilopeganum was analyzed in a systematic context only with other genera of Rutaceae (Townsend, 1986). by Engler (1896, 1931), but its narrow occurrence in the Three At the family level, Rutaceae (161 genera/1815 species; Stevens, Gorges Reservoir area of central China (Song et al., 2004; Tang 2001 onwards, Angiosperm Phylogeny Website) have been investi- et al., 2007) prevented its inclusion in more recent taxonomic gated morphologically (Engler, 1896, 1931; Saunders, 1934; treatments (e.g., Townsend, 1986; Da Silva et al., 1988). Moore, 1936; Scholz, 1964; Tilak and Nene, 1978), molecularly Despite the systematic importance of the above-mentioned (Chase et al., 1999; Scott et al., 2000; Samuel et al., 2001; Morton diagnostic features, relationships and taxonomic boundaries et al., 2003), and biochemically, owing to their remarkable diver- among the six genera of Ruteae (Engler, 1896, 1931) remain con- sity of secondary chemical compounds (Price, 1963; Fish and troversial. Townsend (1986) observed that the states of some char- Waterman, 1973; Waterman 1975, 1983, 1990; Gray and Water- acters traditionally used to differentiate the genera overlap or man, 1978; Waterman and Grundon, 1983; Kong et al., 1986; Ng suggest contradicting sister-group relationships (Table 2). For et al., 1987; Da Silva et al., 1988; Zakaria, 2001). However, different example, the ranges of the number of ovules per locule overlap types of characters led to contrasting systematic conclusions. For across Ruta and allied genera. The presence of cuneate filaments fa- example, some taxonomic groups recognized in the most compre- vors Cneoridium and Thamnosma as sister taxa, whereas spherical hensive morphological study (Engler, 1896, 1931) and the most re- seeds link Cneoridium with Dictamnus. Moreover, Townsend cent chemotaxonomic survey (Da Silva et al., 1988) of Rutaceae (1986) argued that there are no grounds for considering Haplophyl- conflict with each other and with the groups supported in the lum to be more closely related to Ruta than to Thamnosma, as pro- broadest molecular phylogenies available until now (Chase et al., posed by Engler (1896, 1931). In fact, while Ruta and Haplophyllum 1999; Scott et al., 2000). The cited molecular studies, based on share several morphological similarities, including translucent dots sparse character and taxon sampling, supported Ruta either as sis- on the leaves, yellowish flowers, a thick nectary disk, a short thick ter to a genus of subfamily Flindersioideae (Chase et al., 1999), or style, and connate carpels, they can be clearly distinguished by dif- as sister to a clade of subfamily Citroideae (Scott et al., 2000), while ferences in petal margins, flower merism, seed shape, and pollen Engler (1896, 1931) had placed it within subfamily Rutoideae. morphology. Furthermore, Townsend (1986) showed that the pol- In his comprehensive morphological study of Rutaceae Engler len grains of Ruta and Thamnosma are more morphologically simi- (1896, 1931) divided tribe Ruteae into two subtribes: Rutinae, lar to each other than to those of Haplophyllum. comprising Ruta, Thamnosma Torrey and Frémont, Boenninghause- The inclusion of Dictamnus albus L., the only species of the genus nia Reichb. ex Meissner, Cneoridium Hook.f., and Psilopeganum Dictamnus and subtribe Dictamninae, in Ruteae (Engler 1896, Hemsl. ex Forb. and Hemsl.; and Dictamninae, consisting only of 1931; Table 1) is also contentious, for this species is distinct from Dictamnus L. (Table 1). Furthermore, he split Ruta into subgenus all other Rutaceae due to the presence of special quinolones and Euruta Engl., housing five species, three of which were originally limonoids and the absence of coumarins (Da Silva et al., 1988). Fur- described by Linnaeus (1735, 1753), and subgenus Haplophyllum thermore, Moore (1936) remarked that the floral anatomical differ- (around 50 species; see Table 1). Later systematic treatments (Mes- ences between Dictamnus and Ruta are greater than those between ter and Vicol, 1971; Townsend, 1986; Da Silva et al., 1988; Navarro any two genera within any other tribe of Rutaceae, thus criticizing et al., 2004), however, ranked Haplophyllum at the generic level, as the inclusion of Dictamnus and Ruta in Ruteae. Therefore, consider- originally proposed by Jussieu (1825), reducing the number of spe- ing the above-mentioned criticisms towards Engler’s (1896, 1931) cies in Ruta from around 60 to 8, as currently recognized (Town- classification of Ruteae, Townsend (1986) called for a comprehen- send, 1968; Bramwell and Bramwell, 2001). sive systematic re-examination of the entire tribe. The six genera included in Ruteae by Engler (1896, 1931) were Among the genera of Ruteae (Engler, 1896, 1931), Ruta is charac- each distinguished by the following morphological traits (Table 1): terized by strong-smelling ethereal oils in its leaves, greenish-yellow petals with dentate or fimbriate margins, and inflorescences Table 1 with pentamerous terminal flowers and tetramerous lateral flowers Engler’s (1896, 1931) classification of Ruteae, with subsequent modifications by Townsend (1986) and Da Silva et al. (1988) (Townsend, 1968). As currently circumscribed (Townsend, 1968; Bramwell and Bramwell, 2001), Ruta includes eight species of peren- Engler (1896, 1931) Townsend (1986) Da Silva et al. (1988) nial shrubs, with four species widely distributed in the Mediterra- Morphology Morphology Phytochemistry nean (R. chalepensis L., R. graveolens L., R. angustifolia Pers., R. Tribe Ruteae montana (L.) L.), one species endemic to the islands of Corsica and Subtribe Rutinae Ruta-tribe Sardinia (R. corsica DC.), and three species endemic to the Canary Is- Boenninghausenia — Boenninghausenia lands (R. pinnata L.f., R. oreojasme Webb and Berth., Thamnosma Thamnosma Thamnosma Cneoridium — Cneoridium R. microcarpa Svent.). Recently, the populations of R. corsica from Ruta Sardinia have been described as a ninth species, R. lamarmorae, based Subgenus Euruta Ruta Ruta on morphological, karyological, and ecological differences with the Subgenus Haplophyllum Haplophyllum Haplophyllum populations of R. corsica from Corsica (Bacchetta et al., 2006). Psilopeganum —— Overall, morphological data have not been successful in elucidat- Subtribe Dictamninae Dictamnus-tribe ing the relationships and taxonomic boundaries of Ruteae owing to Dictamnus — Dictamnus (i) the paucity of characters diagnostic for the genera within Ruteae, Taxa not treated by the authors are indicated with a dash. (ii) the conflicting and overlapping nature of the characters tradi-

G. Salvo et al. / Molecular Phylogenetics and Evolution xxx (2008) xxx–xxx 3

Table 2 Seven morphological characters, with their respective states, used by Engler (1896, 1931) to discriminate among the genera of tribe Ruteae

Ruta Thamnosma Cneoridium Boenninghausenia Psilopeganum Dictamnus Flower merism 4-Merous or 5-merous 4-Merous 4-Merous 4-Merous 4-Merous 5-Merous Seed shape Angled dorsally Almost reniform Spherical Reniform Reniform Spherical Number of ovules per locule 6–12 5–6 2 6–8 5–6 3 Number of carpels 4 or 5 2 1 4 2 5 Nectary disk shape Cushion-shaped Variable Ring-shaped Cup-shaped Small and narrow Ring-shaped at the end Stigma shape Simple Capitate Almost spherical Simple Capitate Simple Filament shape Broader at base Cuneate Cuneate Filiform n.a. Club-shaped with protruding glands

The states of these characters are either overlapping or suggest different sister-group relationships (see text). n.a., not available. tionally used to establish relationships within Ruteae, and (iii) the sp.), Cneoridium (1/1 sp.), and Dictamnus (1/1 sp.; see Supplemen- different taxonomic value assigned by different authors to compar- tary data 1). It was impossible to sample Psilopeganum sinense, be- ative characters. With respect to Engler’s (1896, 1931) classification cause it is restricted to the Three Gorges Reservoir area, in the Hubei of Ruteae, the most controversial issues are the placement of Haplo- province of central China, and is endangered (Song et al., 2004; Tang phyllum within Ruta and the inclusion of Dictamnus in Ruteae (Town- et al., 2007). Furthermore, this taxon is poorly represented in the send, 1986; Da Silva et al., 1988; see Table 1). herbaria that were visited during the duration of the present study Phytochemical characters have also been used to generate taxo- (i.e., W, LE, P, BR). In order to elucidate relationships within Ruta, nomic treatments of Rutaceae (e.g., Kong et al., 1986; Ng et al., different accessions from the eight species of the genus were se- 1987; Samuel et al., 2001; Zakaria, 2001), even though they pose spe- lected. Five accessions of R. corsica from Corsica and Sardinia were cificproblemsthatappeartolimittheirtaxonomicvalue.Firstly,phy- sampled to verify the treatment of the populations from Sardinia as tochemical information on the family is fragmentary, with only 30% a separate species (i.e., R. lamarmorae; Bacchetta et al., 2006). To of the species of Rutaceae examined (Waterman, 1990). Secondly, test the monophyly of Ruteae (Engler, 1896, 1931), in particular convergence in the production of secondary chemical compounds the inclusion of Dictamnus in the tribe, eight taxa outside the tribe, has been regarded as a primary source of erroneous systematic con- belonging to subfamilies Rutoideae and Toddalioideae, were se- clusions (Hegnauer, 1966; Mothes, 1981; Waterman, 1990; Water- lected. Choice of outgroups was guided by Engler’s (1896, 1931) man, 1998). For example, Waterman and Grundon (1983) argued classification of Rutaceae and previous phylogenetic findings that the synthesis of carbazole, benzophenanthridine, and quinolone (Chase et al., 1999; Scott et al., 2000). Because the molecular phylo- alkaloids, occurring in taxa of Rutaceae with little or no immediate genetic analysis of Scott et al. (2000) placed Ruta as sister to mem- affinity with each other, originated by convergent evolution. bers of subfamily Aurantioideae, rather than Rutoideae, as Considering the controversial interpretation of morphological suggested by Engler (1896, 1931), and because the monophyly of (Townsend, 1986) and biochemical characters (Da Silva et al., the tribes and subfamilies of Rutaceae proposed by Engler (1896, 1988; Waterman, 1990), which produced contradictory taxo- 1931) has been questioned (Da Silva et al., 1988; Chase et al., nomic treatments for Ruteae (Table 1), and the conflict between 1999), taxa from Meliaceae and Simaroubaceae, closely related to traditional taxonomies (Engler, 1896, 1931) and available molec- Rutaceae (Gadek et al., 1996; Muellner et al., in press), were chosen ular phylogenies of Rutaceae (Chase et al., 1999; Scott et al., as outgroups, to reduce the possibility that the rooting taxa might 2000), we performed a detailed phylogenetic study based on se- fall within the ingroup. The final matrix contained 73 accessions: quences from three chloroplast DNA markers to address the fol- 66 belonging to Rutaceae, four to Meliaceae, and three to Simaroub- lowing questions: (1) Does the molecular phylogeny support aceae. Included material, voucher information, sources, and Gen- Engler’s (1896, 1931) circumscription of Ruteae and, specifically, Bank/EBI accession numbers are listed in Supplementary data 1. the inclusion of Dictamnus in the tribe? (2) Does the molecular phylogeny support Engler’s (1896, 1931) circumscription of Ruta 2.2. Character sampling and, specifically, the treatment of Haplophyllum as a subgenus of Ruta? (3) Does the molecular phylogeny support the newly de- After performing preliminary analyses with different cpDNA scribed species R. lamarmorae? (4) Do phylogenetic analyses of markers, three markers that provided sufficient resolution at our le- morphological, phytochemical, and DNA sequence data yield vel of investigation and allowed unequivocal alignments were cho- the same or different relationships among Ruta and closely re- sen: the matK gene, the rpl16 intron, and the trnL–trnF intergenic lated genera? (5) Which morphological and/or phytochemical spacer, which already proved effective at resolving inter-generic characters are congruent with the clades recovered from the relationships in other groups of angiosperms (Simões et al., 2004; molecular phylogenetic analysis? More generally, the discussion Guggisberg et al., 2006; Marazzi et al., 2006; Rutschmann et al., of our results on the phylogeny of Ruteae provides an opportu- 2007). nity to elaborate on the sources of discrepancy among different types of data, one of the fundamental debates in systematics. 2.3. DNA extraction, amplification and sequencing

2. Materials and methods Prior to DNA extraction, silica-dried leaf material (15–20 mg) was ground using glass beads and a MM 3000 shaker (Retsch 2.1. Taxon sampling GmbH, Haan, Germany). Total genomic DNA was extracted using the DNeasy Plant Mini Kit from QIAGEN AG (Basel, Switzerland), Ruta and its most closely related genera (Engler, 1896, 1931; following the manufacturer’s instructions. The matK cpDNA coding Townsend, 1986; Da Silva et al., 1988), with the exception of Psilo- region was amplified using primers 1F and 1R (Sang et al., 1997). peganum (1 sp.), were sampled: Ruta (8/8 species), Haplophyllum The rpl16 intron was amplified using primers F71 and R1516 (24/66 species), Thamnosma (5/9 species), Boenninghausenia (1/1 (Baum et al., 1998). The trnL–trnF spacer was amplified with the

4 G. Salvo et al. / Molecular Phylogenetics and Evolution xxx (2008) xxx–xxx primers e and f (Taberlet et al., 1991). All PCR were 20 ll in vol- The individual partitions and the combined matrix without gap ume. Each reaction included 9.2 ll of ddH2O, 2 ll of Taq Buffer coding were analyzed using maximum parsimony (MP). The com- [10Â, 15 mM MgCl2], 1.6 ll of MgCl2 [25 mM], 3.2 ll of dNTP bined matrix with gap coding was analyzed using both MP and [1.25 mM], 0.2 ll of Taq Polymerase [5 U/ll], 1 ll of BSA, 0.4 ll Bayesian Inference (BI). Parsimony analyses were conducted using of each primer (forward and reverse), and 2 ll of DNA template. PAUPÃ4.0b10 (Swofford, 2001). All changes were treated as unor- Amplification of the matK region consisted of 2 min at 94 °C fol- dered and equally weighted (Fitch, 1971). Tree search was per- lowed by 30 cycles of: 1.5 min denaturation (94 °C), 2 min anneal- formed using the following protocol: (i) a heuristic search was ing (53 °C), and 3 min extension (72 °C). After the last cycle, the carried out with 1000 replicates of random taxon addition se- temperature was kept at 72 °C for the last 15 min of extension quence and 10 trees held at each step, and tree bisection–recon- and then lowered to 4 °C. Amplification of both the rpl16 and nection branch swapping (TBR) on best trees only, with no more trnL–trnF regions consisted of 2 min at 94 °C followed by 35 cycles than 100 trees saved per replicate; (ii) the best trees found in (i) of: 0.5 min denaturation (94 °C), 1 min annealing (52 °C), and were then used as starting trees for a second heuristic search using 1.75 min extension (72 °C). After the last cycle the temperature TBR branch swapping until all swapping options were explored, was kept at 72 °C for 10 min of extension and then lowered to and saving multiple trees (MULTREES option in effect). The STEEP- 4 °C. All PCR and cycle sequencing reactions were run on a TGradi- EST DESCENT option was used in both (i) and (ii). Relative levels of ent thermocycler (Biometra, Goettingen, Germany). In order to homoplasy in all partitions were assessed using the consistency detect amplified DNA target regions and possible contamination, index (CI) and the retention index (RI) as implemented in PCR products were separated on 1% agarose gels, stained with ethi- PAUPÃ4.0b10 (Swofford, 2001). dium bromide, and viewed under UV light. Successfully amplified Relative support for each node obtained by MP was assessed products were purified with the GFX PCR DNA and Gel Band using bootstrap re-sampling (Felsenstein, 1985). The following purification Kit (Bioscience Amersham, Otelfingen, Switzerland), protocol was employed: heuristic search, 1000 bootstrap repli- following the manufacturer’s recommendations. cates, 100 random addition sequence replicates with three trees Cycle sequencing reactions were carried out using the BigDyeTM held at each step, TBR swapping with STEEPEST DESCENT and Terminator Mix (Applied Biosystems, Inc., Foster City, CA) and the saving no more than ten trees per replicate. same primers as above. The sequencing protocol consisted of 24 Bayesian inference was performed in MRBAYES v3.1.2 (Hulsen- cycles of 10 s denaturation (96 °C), 5 s annealing (50 °C), and beck and Ronquist, 2001; Ronquist and Huelsenbeck, 2003), after 4 min elongation (60 °C). Products were run on an ABI 3100 Genet- determining the model of evolution most suitable for each individ- ic Analyzer (Applied Biosystems, Inc., Foster City, CA) according to ual cpDNA region with the Akaike Information Criterion (AIC; the manufacturer’s protocol. For each region both strands were Akaike, 1974) in ModelTest 3.06 (Posada and Crandall, 1998). Sub- sequenced. sequently, the commands ‘‘lset NST = 6, RATES = gamma” and ‘‘lset coding = variable” were entered in MRBAYES v3.1.2 for the nucleo- 2.4. Alignment and phylogenetic analyses tide and gap characters, respectively (Ronquist and Huelsenbeck, 2003). The analysis was performed with four Monte Carlo Markov Sequences were edited and assembled using Sequencher 4.2TM chains (one cold and three incrementally heated) run for 5 Â 106 software (Gene Codes Corp., Ann Arbor, MI, USA). Base positions generations, with trees sampled every 1000th generation (NGEN = were individually double-checked for agreement between the 1.000.000, PRINTFREQ = 1000, SAMPLEFREQ = 1000, NCHAINS = 4). complementary strands. All sequences were visually aligned in Each chain used a random tree as starting point and the default MacClade 4.06 (Maddison and Maddison, 2000). Regions of ambig- temperature parameter value of 0.2. Two independent Markov uous alignment were excluded from the analysis (Kelchner, 2000). chain Monte Carlo runs were carried out to check for convergence Gap positions were treated as missing data, unequivocally aligned on the same region of tree space. The first 25,000 sampled trees gaps being coded as presence/absence of characters with the soft- were discarded as ‘‘burn in” after checking for stability on the ware GapCoder (Young and Healy, 2003) and then added as binary log-likelihood curves. The remaining trees were imported into characters to the data matrix. PAUPÃ4.0b10 (Swofford, 2001) and used to build a 95% majority Three data partitions were defined, corresponding to the three rule consensus tree showing the posterior probabilities (PP) of all loci of the chloroplast genome examined in this study. The individ- observed bi-partitions. ual partitions were initially analyzed separately to establish whether there were any strongly supported (i.e., >85 bootstrap 2.5. Analyses of constrained topologies percentage, BP), incongruent clades among the respective trees. Since no such incongruence was detected (see Section 3), the se- Engler’s (1896, 1931) classification of tribe Ruteae has been crit- quences of the three loci were combined in a single dataset. The icized with respect to two points: the placement of Dictamnus in combined matrix was then analyzed phylogenetically either with the tribe (Moore, 1936; Da Silva et al., 1988) and the treatment the gaps treated as missing data (‘‘combined without gap coding”) of Haplophyllum as a subgenus of Ruta (Townsend, 1986). Two or with the gaps coded in GapCoder (Young and Healy, 2003; topological constraints were thus defined: (i) Dictamnus within Ru- ‘‘combined with gap coding”). teae and (ii) Haplophyllum as sister to Ruta. For each constraint, a

Table 3 Results of S–H tests on two topological constraints

Topologies No. of MP trees MP length ÀlnL scoresa ÀlnL differencesb p Valuesc Unconstrained 6 2092 17570.505 —— Constraint (i): Dictamnus within Ruteae 94 2388 17626.010 55.506 0.080 Constraint (ii): Haplophyllum sister to Ruta 940 2455 17784.520 214.025 <0.001

a Highest likelihood scores assigned to MP trees; the highest likelihood score of the unconstrained trees (in boldface) was used for comparisons with the likelihood scores of all constrained trees. b Differences between likelihood scores of constrained and uncontrained trees. Only the differences between the unconstrained tree with highest ÀlnL score and the constrained trees with highest ÀlnL scores are shown. c p Values associated with differences between likelihood scores (signiﬁcant values in boldface).

G. Salvo et al. / Molecular Phylogenetics and Evolution xxx (2008) xxx–xxx 5 heuristic search was performed to find the shortest trees, which to build a matrix (see Supplementary data 2). The number of were then assigned a likelihood score (Table 3) using the parame- ovules per locule was not included in the matrix of Supplementary ters previously estimated (see Table 4). The likelihood scores of the data 2, because states overlapped extensively. Phytochemical char- 94 MP trees with constraint (i) and of the 940 MP trees with con- acters were selected from Da Silva et al. (1988), the most recent straint (ii) were each compared with the highest likelihood score and comprehensive chemotaxonomic survey of Rutaceae. These of the MP unconstrained trees, for a total of 1034 pair-wise com- characters referred to the presence or absence of specific phyto- parisons, and the differences between likelihood scores calculated chemical compounds and consequently were scored as binary data (Shimodaira–Hasegawa (S–H) test; Shimodaira and Hasegawa, (Supplementary data 3). The molecular dataset was built by keep- 1999; Goldman et al., 2000; Lee and Hugall, 2003). Significance lev- ing only one, randomly chosen exemplar sequence from the com- els for the differences were checked on a distribution generated by bined matrix with coded gaps for each of the six genera. The using a RELL technique (resampling estimated log likelihoods; selected sequences (indicated by an asterisk in Supplementary Kishino and Hasegawa, 1989). The S–H tests were performed in data 1) were visually re-aligned in MacClade 4.06 (Maddison and PAUPÃ4.0b10 (Swofford, 2001) on the matrix without gap coding, Maddison, 2000). A global dataset was also constructed by combin- because this software does not allow for likelihood estimation in ing all types of characters. For each of the four datasets, exhaustive datasets containing two or more partitions explained by different searches were carried out using maximum parsimony with the models of evolution. program PAUPÃ4.0b10 (Swofford, 2001). Branch support was calculated with 10,000 bootstrap replicates using a Branch and Bound 2.6. Phylogenetic comparisons among different datasets search strategy and an ‘‘as is” taxon addition sequence. The trees resulting from the molecular, morphological, and phy- To evaluate whether different types of data produce congruent tochemical datasets were compared with each other in two ways. phylogenies, we generated comparable matrices from morphology, First, because phylogenetic incongruence should only include cases biochemistry, and DNA sequences. Published information was where conflicting clades are strongly supported, topologies were available mainly at the genus level for the morphology of Rutaceae evaluated by direct node-to-node comparisons using branch sup- (Engler, 1896, 1931; Saunders, 1934; Scholz, 1964), and at the spe- port values (e.g., Mason-Gamer and Kellogg, 1996; Graham et al., cies level for the secondary chemical compounds (e.g., Price, 1963; 1998). Node-to-node comparisons were executed by using the ta- Fish and Waterman, 1973; Waterman 1975, 1983, 1990; Gray and ble of ‘‘bipartitions found in one or more trees and frequency of Waterman, 1978; Waterman and Grundon, 1983; Kong et al., occurrence” from the bootstrap output produced in PAUPÃ4.0b10 1986; Da Silva et al., 1988), but not for all the species included in (Swofford, 2001). Secondly, the trees were compared by means of our molecular analyses. Therefore, comparisons among different the incongruence length difference (ILD) test (Farris et al., 1994), datasets were performed at the genus level and only on Haplophyl- implemented as the ‘‘partition homogeneity” test in PAUPÃ4.0b10 lum, Thamnosma, Boenninghausenia, Cneoridium, and Ruta, because (Swofford, 2001). For the ILD test, 1000 random repartitions were the results of our molecular analyses excluded Dictamnus from Ru- used and a branch and bound tree search was implemented. Fol- teae, in agreement with Moore (1936) and Da Silva et al. (1988) lowing suggestions by Cunningham (1997) and Lee (2001), the test (see Section 3 below). The genus Choisya, belonging to the subfam- was carried out after removing uninformative characters. ily Rutoideae (Engler, 1896, 1931), was chosen to root the resulting trees, because it is closely related to Ruteae (Da Silva et al., 1988) 2.7. Character mapping and both morphological and phytochemical data were available for it. We investigated whether any of the morphological and/or phy- Morphological characters that vary among the six genera listed tochemical characters listed in Supplementary data 2 and 3 were above were selected from descriptions in Engler (1931) and Town- congruent with the relationships inferred from the analysis of send (1986), scored for different states (i.e., multistate), and used the 6-taxon molecular dataset. Non-molecular characters were

Table 4 Character and tree diagnostics, substitution models, and parameters estimated by ModelTest for the ﬁve data partitions

matK rpl16 trnL–trnF Combined without gap coding Combined with gap coding Aligned length 1564 1321 709 3594 3798 Parsimony informative ntps (% of aligned ntps) 429 (27.4) 294 (22.3) 125 (17.6) 848 (23.6) 957 (25.2) No. of trees 600 803 70 6 16 No. of steps 1076 723 274 2092 2367 CI (CI ex) 0.744 0.765 0.803 0.752 0.751 (0.682) (0.702) (0.758) (0.691) (0.682) RI 0.950 0.947 0.960 0.948 0.947 Model selected TVM + G TVM + G TVM + G TVM + G TVM + G & binary model ÀlnL 8589.0039 6065.5581 2608.8774 17638.0703 — Freq. [A] 0.2840 0.4054 0.4025 0.3525 — Freq. [C] 0.1722 0.1450 0.1564 0.1575 — Freq. [G] 0.1690 0.1522 0.1651 0.1658 — Freq. [T] 0.3749 0.2974 0.2759 0.3242 — R.r.o.s. [A-C] 1.1585 1.4023 0.5307 1.1031 — R.r.o.s. [A-G] 1.5442 1.3557 0.8744 1.2461 — R.r.o.s. [A-T] 0.2248 0.2041 0.0938 0.1931 — R.r.o.s. [C-G] 0.8399 0.7504 0.6037 0.7632 — R.r.o.s. [C-T] 1.5442 1.3557 0.8744 1.4942 — R.r.o.s. [G-T] 1 1 1 1 — I0000— C 0.8744 0.9871 0.5913 0.8191 —

For the ‘‘combined with gap coding” partition parameter values are not present because they were re-estimated by MrBayes. Ntps, nucleotide positions; CI, consistency index; CI ex, consistency index excluding parsimony uninformative characters; RI, retention index; Freq., frequency; R.r.o.s., relative rate of substitution; I, proportion of invariable sites; C, gamma distribution shape parameter.

6 G. Salvo et al. / Molecular Phylogenetics and Evolution xxx (2008) xxx–xxx mapped on the molecular topology using maximum parsimony found, respectively, as compared to the 6 MP trees of 2092 steps and both accelerated (ACCTRAN) and delayed (DELTRAN) character resulting from the unconstrained search (Table 3). All pair-wise state optimizations, as implemented in MacClade 4.06 (Maddison comparisons involving constraint (ii) produced significant differ- and Maddison, 2000). ences between likelihood scores (p < 0.001; Table 3, in bold). How- ever, none of the comparisons involving constraint (i) produced significant results (p values between 0.080 and 0.038; Table 3). 3. Results 3.3. Comparisons among different datasets 3.1. Alignment and phylogenetic analyses For each six-taxon dataset, only one MP tree was found (Fig. 2; Aligned lengths, character and tree statistics, CI and RI values Table 5). The inter-generic relationships inferred from the six-tax- for all five partitions are summarized in Table 4. Forty-six and 27 on molecular dataset (Fig. 2a) were identical to those generated nucleotide positions were excluded from the rpl16 and trnL–trnF from the 73-accession matrix (Fig. 1), with Ruta sister to Boennin- partitions, respectively, owing to ambiguities in the alignment ghausenia/Thamnosma and these three genera, in turn, sister to caused by strings of mononucleotides (Kelchner, 2000). Among Haplophyllum/Cneoridium, indicating that random selection of one the three cpDNA partitions, the matK dataset contained the highest exemplar per genus did not change the phylogenetic pattern. All proportion of parsimony-informative characters (27.4%) and pro- the nodes of the molecular tree were highly supported (Fig. 2a), duced the best resolved tree, whereas the trnL–trnF partition had whereas only one node each received high BP value in the morpho- the highest CI and RI values (Table 4). Because no strongly sup- logical (Fig. 2b) and molecular trees (Fig. 2c). The relationships of ported (>85 BP) incongruent clades were detected among individ- the global tree were identical to those of the molecular tree and ual trees, the three partitions were combined, producing an strongly supported (Fig. 2d). The Boenninghausenia/Thamnosma alignment of 3594 characters. The gaps of the combined matrix clade received maximum support (100 BP) from the molecular data yielded 204 additional characters, for a total of 3798 characters and was found in 86% of the bootstrap replicates of the morpholog- (Table 4). The combined matrix with gaps coded was used for final ical data. The Haplophyllum/Cneoridium clade, with 99 BP in the MP and Bayesian analyses, because it had a higher number of par- molecular tree, was found in none of the bootstrap replicates of simony informative sites and similar CI and RI values as compared the non-molecular datasets. The phytochemical tree shared no with the combined matrix without coded gaps (see Table 4). clades with the molecular tree and its single strongly supported For all three DNA regions the AIC of ModelTest selected the clade (i.e., Ruta/Haplophyllum; 97 BP) received low bootstrap sup- same model of evolution, TVM + G, which was implemented in port (66 BP) in the morphological tree. the Bayesian analysis, with parameters estimated again from the Based on the results of the ILD test, the molecular dataset was data. The two runs of the Bayesian analysis produced identical found to be significantly incongruent with both the morphological 50% majority-rule consensus trees, suggesting convergence on (p = 0.002) and phytochemical (p = 0.0001) datasets, while the the same region of tree space. Each run reached a stationary likeli- morphological and phytochemical datasets were not significantly hood after approximately 280.000 generations, which were not incongruent with each other (p = 0.41). used to build the Bayesian consensus tree. The 95% majority-rule consensus tree obtained from one run of the Bayesian analysis, 3.4. Character mapping with posterior probabilities (PP), and bootstrap percentages (BP) obtained by bootstrapping the same matrix under parsimony, is Sixteen of the 47 non-molecular characters mapped on the shown in Fig. 1. This tree was similar to the strict consensus tree molecular tree exhibited equivocal reconstructions, that is, charac- found from the MP search of the same matrix (not shown), except ter-state transitions occurred in different branches depending on that in the MP tree some groups within Haplophyllum were re- whether ACCTRAN or DELTRAN were used for the optimization. solved differently compared to the Bayesian tree. The remaining 31 characters were optimized unequivocally and The main phylogenetic results are the following (Fig. 1): (1) three of them changed uniquely along the branches of the simpli- Ruta, Boenninghausenia, Thamnosma, Haplophyllum, and Cneoridium fied molecular tree (Fig. 3). Character 9 switched from a short and form a monophyletic group with maximum support (100 BP and thick to a long and thin style along the branch leading to Tham- 1.00 PP); (2) Dictamnus forms a clade with Skimmia Thunb. and Ori- nosma/Boenninghausenia, character 11 switched from introrse to xa Thunb. (100 BP and 1.00 PP); (3) the eight species currently as- slightly-introrse anther opening along the same branch, and char- cribed to Ruta (Townsend, 1968; Bramwell and Bramwell, 2001) acter 34 switched from the absence to the presence of acridones of form a strongly supported clade (100 BP and 1.00 PP) that is sister type H1 along the branch subtending Ruta/Thamnosma/Boennin- to a clade consisting of Thamnosma and Boenninghausenia (100 BP ghausenia (see also Supplementary data 2 and 3). In contrast, a and 1.00 PP), while Haplophyllum is sister to Cneoridium (100 BP transition from obovate to ovate petals (character 5) and a switch and 1.00 PP); (4) R. chalepensis and R. angustifolia form a clade sis- from the absence to the presence of 2-quinolones of type G1.1 ter to R. graveolens and these three species are sister to R. corsica (character 23) occurred more than once, but nonetheless sup- (clade I); (5) the three Ruta species from the Canary Islands (R. pin- ported the clades Haplophyllum/Cneoridium and Thamnosma/Boen- nata, R. microcarpa, and R. oreojasme) form a strongly supported ninghausenia, respectively (Fig. 3). clade (100 BP and 1.00 PP; clade III); (6) the relationship between R. montana (clade II; Fig. 1) and clades I and III described above re- 4. Discussion mains unresolved; (7) within R. corsica there is a strongly supported split between the samples from Sardinia (86 BP and 4.1. Circumscription of Ruteae and Ruta 1.00 PP) and those from Corsica (99 BP and 1.00 PP). In the most comprehensive classification of Rutaceae, Engler 3.2. Analyses on constrained topologies (1896, 1931) proposed the inclusion of Dictamnus in Ruteae. How- ever, the inferred cpDNA phylogeny strongly supportsthe monophyly When Dictamnus was forced inside Ruteae (constraint i) and of a clade formed byRuta, Boenninghausenia, Thamnosma, Haplophyl- when Haplophyllum was constrained to be sister to Ruta (constraint lum and Cneoridium,whileDictamnus is embedded in a clade compris- ii), 94 MP trees of 2388 steps and 940 MP trees of 2455 steps were ing members of Zantoxyleae, Diosmeae and Toddalioideae (Fig. 1).

G. Salvo et al. / Molecular Phylogenetics and Evolution xxx (2008) xxx–xxx 7

Ruta chalepensis Ruta chalepensis 1.00 Ruta chalepensis 100 Ruta chalepensis Ruta chalepensis 0.98 Ruta chalepensis 73 Ruta angustifolia 1.00 0.95 Ruta angustifolia 100 Ruta angustifolia 71 Ruta angustifolia I 1.00 Ruta graveolens 1.00 1.00 97 Ruta graveolens 89 100 Ruta graveolens 1.00 Ruta corsica Sardinia 1.00 86 Ruta corsica Sardinia 100 Ruta corsica Corsica 1.00 Ruta corsica Corsica 1.00 II 99 Ruta corsica Corsica 100 1.00 Ruta montana 100 Ruta montana 1.00 Ruta pinnata 1.00 III 1.00 86 Ruta pinnata 100 1.00 100 1.00 Ruta microcarpa R 100 87 Ruta microcarpa Ruta oreojasme u Boenninghausenia albiflora 1.00 1.00 Thamnosma hirschii 100 100 1.00 Thamnosma socotrana t 100 Thamnosma montana 1.00 Thamnosma pailensis 100 Thamnosma texanum e 1.00 Haplophyllum acutifolium 79 Haplophyllum acutifolium 1.00 Haplophyllum acutifolium a 100 Haplophyllum robustum 1.00 Haplophyllum versicolor e 100 1.00 Haplophyllum canaliculatum 100 Haplophyllum virgatum Haplophyllum stapfianum 1.00 Haplophyllum linifolium 99 Haplophyllum bastentanum 1.00 Haplophyllum coronatum 1.00 79 Haplophyllum patavinum 100 1.00 0.98 Haplophyllum suaveolens 100 56 1.00 Haplophyllum laeviusculum 1.00 89 Haplophyllum tuberculatum 1.00 88 Haplophyllum sp. 100 Haplophyllum buxbaumii 1.00 Haplophyllum villosum 100 1.00 Haplophyllum erythraeum 1.00 100 Haplophyllum glaberrimum 100 Haplophyllum furfuraceum Haplophyllum latifolium Haplophyllum obtusifolium Haplophyllum dauricum Cneoridium dumosum 1.00 Zanthoxylum simulans 1.00 100 Zanthoxylum americanum 100 Evodia hupehensis Zanthoxyleae 1.00 Choisya ternata 98 Calodendrum capense Diosmeae 1.00 Melicope sp. 100 1.00 Skimmia japonica Toddalieae 1.00 91 Orixa japonica 100 Dictamnus albus Zurich Dictamnus albus Bosnia 1.00 Ailanthus altissima* 1.00 100 Ailanthus giraldii* 100 Simarouba glauca* Swietenia macrophylla* 1.00 1.00 Azadirachta indica* 100 1.00 100 Melia azedarach* 85 Cipadessa baccifera*

Fig. 1. Ninety ﬁve percent majority-rule consensus tree obtained from the Bayesian analysis. Posterior probabilities (PPs) and Bootstrap percentages (BPs; above 50%) are shown above and below branches, respectively. The white bar indicates members of tribe Ruteae sensu Engler (1896, 1931). The gray bar indicates members of tribe Zanthoxyleae. Tribes Ruteae, Zanthoxyleae, and Diosmeae belong to subfamily Rutoideae; tribe Toddalieae belongs to subfamily Toddalioideae. Taxa with an Ã were included in the outgroup.

Molecular evidence thus apparently corroborates Da Silva’s (1988) At least morphologically, Dictamnus can be viewed as an aber- interpretation of tribal boundaries (see Table 1). rant form of uncertain phylogenetic placement, for it shows some The five genera sharing a single origin in the molecular tree (i.e., morphological features that cannot be readily linked with any Ruta, Boenninghausenia, Thamnosma, Haplophyllum and Cneoridium; other taxa of Rutaceae (Moore, 1936). Within Ruteae (Engler, Ruteae s.s. from now on; see Fig. 1) share a number of morpholog- 1896, 1931), Dictamnus differs from other genera in seed structure ical and phytochemical traits, including: the presence of actino- (Corner, 1976) and floral morphology, with large, zygomorphic, morphic, creamy-white to bright-yellow flowers (Engler 1896, white to purple flowers, lanceolate petals, and unusual oil glands 1931); the highest levels of lignans of the aryltetrahydronaphtha- protruding from the carpel walls and the style (Moore, 1936; lene type in Rutaceae (Waterman, 1983; Da Silva et al., 1988); spe- Gut, 1966). The secondary chemistry of the genus is also unique cific classes of coumarins and acridones (Waterman, 1975); and, within Rutaceae. Dictamnus has limonoids, instead of coumarins, uniquely in Rutaceae, the biosynthetic pathway for acridones de- and special quinolones (Da Silva et al., 1988). Furthermore, early void of an oxygen substituent at the C-3 position and also, in some serodiagnostic studies on Rutaceae (Bärner, 1927) showed that cases, at the C-1 position (Waterman, 1983). Therefore, these gen- ‘‘between Ruta chalepensis and Dictamnus albus the reaction was era appear to form a cohesive taxonomic group. only slightly positive, an observation strictly in accord with floral

8 G. Salvo et al. / Molecular Phylogenetics and Evolution xxx (2008) xxx–xxx

a Ruta b Ruta 66 100 0/97 12/0 Boenninghausenia Haplophyllum 100 86/14 Thamnosma Boenninghausenia 86 Haplophyllum Thamnosma 66 99 0/28 100/14 0/0 Cneoridium Cneoridium

Choisya Choisya Molecular Morphological

c Haplophyllum d Ruta 97 Ruta Boenninghausenia 58 0/66 0/0 100 Boenninghausenia 100 Thamnosma

Cneoridium Haplophyllum 98 Thamnosma Cneoridium

Choisya Choisya Phytochemical Global

Fig. 2. Trees obtained from parsimony analyses of molecular (a), morphological (b), phytochemical (c), and global (d) datasets. Choisya served as the rooting taxon. Bootstrap percentages (BPs) generated by the same dataset used to infer the tree are reported above the branches; BPs generated by the rival datasets are reported below the branches: (a) BP morphology/BP phytochemistry, (b) BP molecules/BP phytochemistry, (c) BP molecules/BP morphology.

Table 5 ter case only trees where Haplophyllum is sister to Ruta are Character diagnostics and trees resulting from the analysis of the molecular, allowed. Therefore, it is difficult to compare the significance values morphological, phytochemical, and global datasets of intrinsically different constraints. To our knowledge, the poten- Dataset No. of Parsimony No. of No. of CI RI tial influence of the stringency of the constraint on the significance characters informative optimal steps (CI ex) levels estimated by the S–H test has not yet been investigated. characters trees Hence, while the results of the S–H test do not reject the possible Molecular 3176 179 1 673 0.924 (0.801) 0.791 inclusion of Dictamnus in Ruteae, the optimal cpDNA tree topology, Morphological 16 12 1 41 0.878 (0.828) 0.667 morphological observations (Moore, 1936; Gut, 1966; Corner, Phytochemical 31 18 1 38 0.816 (0.667) 0.667 Global 3223 209 1 769 0.899 (0.760) 0.721 1976), and phytochemical data (Da Silva et al., 1988) all suggest that this genus may not have evolved from the same common CI, consistency index; CI ex, consistency index excluding parsimony uninformative ancestor as the other members of Ruteae (Engler, 1896, 1931). characters; RI, retention index. In Engler’s (1896, 1931) classification, Ruta comprised around 60 species, subdivided in subgenus Euruta, with five species, and subgenus Haplophyllum, with about 50 species (Table 1). If Engler’s inter- anatomy, but disagreeing with the taxonomists’ assignment of pretation were correct, species ascribed to the two subgenera Ruta and Dictamnus near to one another.” (Moore, 1936: 322) would be expected to form sister clades in a phylogeny. However, Despite the morphological and phytochemical differences be- this is not the case in the cpDNA tree inferred in this study (Fig. 1), tween Dictamnus and the other genera of Ruteae and its distant for the eight currently-recognized species of Ruta (R. chalepensis, R. relationship with other Ruteae in the cpDNA phylogeny (Fig. 1), angustifolia, R. graveolens, R. corsica, R. montana, R. pinnata, R. micro- the S–H test suggests that this phylogenetic result is not signifi- carpa, and R. oreojasme Townsend, 1968; Bramwell and Bramwell, cantly different than the inclusion of Dictamnus in Ruteae (Table 2001); form a strongly supported monophyletic group that is sister 3, constraint i). This result contrasts with the significant rejection to Boenninghausenia/Thamnosma, while the 24 species of Haplophyl- of the second constraint assessed by the S–H test, i.e., forcing lum constitute the sister clade of Cneoridium. Moreover, when Haplo- Haplophyllum to be sister to Ruta (see below). However, it should phyllum was constrained to be sister to Ruta, the differences between be noted that forcing Dictamnus in Ruteae applies a rather relaxed the likelihood score of the constrained and unconstrained topolo- constraint compared to forcing Haplophyllum to be sister to Ruta, gies, evaluated by means of the S–H test, were statistically significant because in the former case all the trees with all possible place- (Table 3). Hence, the molecular results corroborate Townsend’s ments of Dictamnus within Ruteae are allowed, whereas in the lat- (1986) interpretation of generic boundaries (Table 1).

G. Salvo et al. / Molecular Phylogenetics and Evolution xxx (2008) xxx–xxx 9

R 34 34 acridones of type H1 5 ovate petals

11 B

5 9 23 T 11 slightly introrse anther opening

9 thin and long style

V H V

23 23 2-quinolones of type G1.1 5 C

Fig. 3. The ﬁve derived non-molecular character states consistent with the clades supported by molecular data: 5, ovate petals (taken from Townsend, 1986); 9, thin and long style (taken from Engler, 1931); 11, slightly introrse anther opening (shown in cross-sectional view with the lower part facing the gynoecium and the arrows indicating the direction of pollen release; personal observation and Prof. P. Endress pers. comm.); 23, 2-quinolones of type G1.1 (taken from Da Silva et al., 1988); 34, acridones of type H1 (taken from Da Silva et al., 1988). Black and white bars indicate non-homoplasious and homoplasious transitions, respectively. R, Ruta; B, Boennighausenia; T, Thamnosma; H, Haplophyllum; C, Cneoridium, O, Outgroup (Choisya). See Supplementary data 2 and 3 for further details.

Townsend (1986) identified morphological differences between Corsican populations of R. corsica (i.e., R. corsica s. str.) by morpho- Ruta and Haplophyllum that had been overlooked by Engler, 1896, logical, ecological and karyological differences (Bacchetta et al., 1931 (Table 2). In fact, the petal margins of Ruta are dentate or fim- 2006). R. corsica s. str. is diploid, has smaller flowers, stamens, briate, whereas those of Haplophyllum are more or less entire; in and ovaries, occurs across a wider altitudinal range (1000–1900 Ruta the lateral flowers are 4-merous and the terminal flowers m.a.s.l.) and its pollen matures in June. R. lamarmorae is tetraploid, are 5-merous, whereas in Haplophyllum both lateral and terminal has bigger flowers, stamens and ovaries, occurs in a more restricted flowers have the same merism (usually 5-merous); the seeds of altitudinal range (1500–1750 m.a.s.l.), and its pollen matures in Ruta are bluntly- to sharply-angled dorsally, whereas they are reni- May. In the molecular phylogeny (Fig. 1), the accessions from Cor- form and dorsally-rounded in Haplophyllum; and finally the pollen sica and Sardinia formed two strongly supported clades, rather grains of Ruta have elongate costae and a finely reticulate to perfo- than being interspersed, thus suggesting that the treatment of rate tectum ornamentation, whereas the pollen grains of Haplo- R. lamarmorae as a separate species might be warranted. The phy- phyllum have thick costae and a closed-striate tectum. logenetic separation between R. corsica s. str. and R. lamarmorae Phytochemically, while Haplophyllum has a predominance of alka- reflects the comparatively high absolute number (seven; data not loids over coumarins (45%), all its most closely related genera have shown) of nucleotide substitutions between the respective acces- either more coumarins (Ruta, 86%; Thamnosma, 88%; Boenningha- sions from Corsica and Sardinia, two islands in close proximity to usenia, 90%) or exclusively coumarins (Cneoridium), an observation each other. In contrast, the accessions of R. chalepensis from distant that, combined with additional phytochemical evidence, led Da Sil- locations (Sicily, Greece, Corsica, Sardinia, mainland France; see va et al. (1988) to recommend that Haplophyllum be recognized as Supplementary data 1) are separated at most by five nucleotide a distinct genus from Ruta. Therefore, our molecular phylogenetic substitutions (data not shown). A more definitive assessment of results, statistical tests on constrained topologies, Townsend’s the proposed specific rank of R. lamarmorae (Bacchetta et al., (1986) detailed morphological analyses, and Da Silva et al. (1988) 2006) must await further evidence from the nuclear genome, phytochemical data all suggest that Haplophyllum should be trea- molecular dating analyses, and inter-fertility studies. ted as a distinct genus, rather than as a subgenus of Ruta. Within Ruta, the seven species represented by multiple acces- 4.2. Comparisons among different datasets sions were all supported as monophyletic by the cpDNA genome and the main clades are congruent with morphological observa- Phylogenies of Ruta and related genera inferred from six-taxon tions or distribution (see Fig. 1). For example, San Miguel (2003) morphological, phytochemical, and molecular matrices were com- stated that R. angustifolia, R. chalepensis, and R. graveolens are mor- pared to identify and localize incongruence between the datasets phologically very similar and virtually impossible to differentiate (Fig. 2), rather than to argue for or against combining data (e.g., in their vegetative parts, whereas R. montana is distinguished by Cannatella et al., 1998). Nodes with low statistical support ambig- its narrower leaves. The three species of Ruta from the Canary Is- uously represent hierarchical patterns within individual datasets, lands are distinct from the remaining species of the genus by being thus conflict among datasets cannot be inferred from comparisons taller (Townsend, 1968; Bramwell and Bramwell, 2001) and having involving weak nodes (de Queiroz, 1993; Mason-Gamer and Kel- larger leaves (G. Salvo, personal observation), consistent with the logg, 1996; Graham et al., 1998; Van der Niet et al., 2005). Follow- observation that insular species exhibit trends toward larger size ing this logic, direct node-to-node comparisons of bootstrap values (Lomolino et al., 2006). detected no incongruent relationships between the molecular and Recently, the populations of R. corsica from Sardinia were de- morphological trees (Fig. 2a and b), while the ILD test suggested scribed as a new species, R. lamarmorae, distinguished from the significant incongruence. Beyond the known criticisms of the ILD

10 G. Salvo et al. / Molecular Phylogenetics and Evolution xxx (2008) xxx–xxx test (e.g., Dolphin et al., 2000; Yoder et al., 2001; Darlu and Lecoin- the species for which certain compounds were sought but were not tre, 2002; Quicke et al., 2007), in the case of the molecular and detected, a problem also encountered in an angiosperm-wide morphological trees of Ruta and related genera, the incongruence study (Nandi et al., 1998). Therefore, the absence of a specific com- estimated by the ILD test might reflect sampling bias in the smaller pound from certain species in a phytochemical data matrix may morphological dataset (12 informative characters, as compared to indicate true absence, lack of detection due to technical limitations, the 179 informative characters of the molecular dataset; Table 5), or missing information. Hence, despite the considerable diversity rather than conflicting phylogenetic signals (e.g., Cannatella et al., of secondary chemical compounds in Rutaceae (e.g., Price, 1963; 1998; Graham et al., 1998). The only significant inconsistencies Waterman, 1975, 1983, 1990; Kong et al., 1986; Da Silva et al., among the three trees (Fig. 2a–c) involved the relationships of Ruta 1988; Samuel et al., 2001; Zakaria, 2001), their systematic and and Haplophyllum, for the two genera were sister to each other in phylogenetic value appears to be fundamentally flawed by homo- the phytochemical tree (BP 97), but in the molecular tree the plasy and methodological issues. Considering the above-men- former was sister to Boenninghausenia/Thamonosma (BP 100) tioned problems, Waterman (1998: 547), the foremost expert on and the latter to Cneoridium (BP 99). What could cause the the secondary compounds of Rutaceae (Fish and Waterman, observed incongruence between the molecular and phytochemical 1973; Waterman, 1975, 1983, 1990; Gray and Waterman, 1978; topologies? Waterman and Grundon, 1983), asserted that ‘‘chemical systemat- In the phylogeny generated from parsimony analysis of the six- ics remains as much an art as a science, and the most appropriate taxon cpDNA matrix (Fig. 2a), the terminal branches subtending use of chemical data appears to be to test phylogenies that have Haplophyllum and Cneoridium are the longest, with 147 and 85 arisen from the interpretation of more complete non-chemical steps, respectively, while the branch leading to their common datasets.” ancestor is much shorter (38 steps; results not shown). Since parsimony methods are known to be particularly vulnerable to long- 4.3. Character mapping analysis and genus diagnosis branch attraction (LBA), as compared to model-based methods (Felsenstein, 1978; Hendy and Penny, 1989; Lewis, 2001), it is pos- The choice of the characters used to build the phylogenies that sible that the Haplophyllum/Cneoridium clade in the molecular tree are in turn utilized to analyze the evolution of selected character (Fig. 2a) be a product of LBA. However, the analysis of the 73-acces- sets remains a controversial issue, because, at its core, it influ- sion molecular matrix by either parsimony or Bayesian methods ences assessments of homology and homoplasy in different data- (Fig. 1) supported the sister relationship between Haplophyllum sets (Brooks, 1996). Considering the dearth, overlapping nature, and Cneoridium, thus suggesting that LBA should not have biased and conflicting taxonomic value of the morphological characters our results for the six-taxon dataset. traditionally used for Ruteae classifications (Townsend, 1986) Homoplasy is often invoked to explain disagreements among and the well-known problem of convergence in the phytochemi- phylogenies inferred from different types of data (e.g., Sanderson cal data of Rutaceae (Waterman, 1990), it seems reasonable to and Hufford, 1996; Wiens et al., 2003). In secondary chemical com- use the strongly supported topology of the molecular tree (Figs. pounds, homoplasy has been repeatedly documented, because 1 and 2a) for the mapping of non-molecular characters (Figs. 3 similar selective pressures can lead to the evolution of pathways and 4). producing similar end-products in unrelated taxa (Price, 1963). The optimization of character-state transitions on the molecular Within Rutaceae, the expression of coumarin prenylation patterns, topology underscored the difficulty of finding non-molecular syn- furoquinoline and acridone oxygenation patterns, and the develop- apomorphies that are consistent with the clades of the molecular ment of the acridone and carbazole nuclei are all known to occur in tree. Out of 47 mapped characters, only five provided character- unrelated taxa (Waterman, 1990). Most of the secondary chemical state transitions that supported the molecular clades (characters compounds used to infer the phytochemical phylogeny of Fig. 2c 5, 9, 11, 23, 34; Fig. 3). Moreover, the morphological characters are alkaloids. Some studies have shown that alkaloid biosynthesis used by Engler (1896, 1931) to differentiate among the genera of in plants is both plastic and labile, for the responsible genes can be Ruteae (i.e., characters 1, 2, 3, 4, 7, 8, 12, 14; Supplementary data repeatedly switched on and off during development and across 2) were either equivocally reconstructed or inconsistent with the evolutionary times (McKey, 1980; Wink and Witte, 1983; Water- molecular clades. Engler (1896, 1931) placed high taxonomic man, 1998; Wink, 2003). For example, the genes responsible for importance on flower merism (Table 2). However, the numbers the biosynthesis of quinolizidine alkaloids are widely represented of sepals, petals, stamens, and carpels are polymorphic in Ruta in the plant kingdom, but are only expressed in a few unrelated and the latter also in Haplophyllum (characters 1–4, see Supple- families (Wink and Witte, 1983). Similar examples include the evo- mentary data 2; Saunders, 1934; Moore, 1936). Additionally, all lution of the benzylisoquinoline alkaloid biosynthesis (Liscombe four characters were equivocally reconstructed (Fig. 4). Conse- et al., 2005), and the production of pyrrolizidine alkaloids as a de- quently, the change between 5 and 4 elements in the perianth fense against herbivores (Reimann et al., 2004). Waterman (1975) whorls and between 10 and 8 stamens in the androecium could commented that the alkaloid types found in Rutaceae, with the have occurred either by reduction or by expansion (Fig. 4), corrob- exception of the canthinones and carbazoles, exhibit a highly ran- orating the suggestion that these characters may be evolutionarily dom distribution and fail to support any of the taxonomic groups labile (Endress, 1990, 1999), and thus of limited taxonomic value. proposed by Engler (1896, 1931). Therefore, considering the Conversely, in the gynoecium, the transition to two carpels in well-known problem of convergence among secondary compounds Thamnosma and one in Cneoridium always occurred by reduction, (Hegnauer, 1966; Mothes, 1981), it seems more likely that the regardless of the ancestral state for the clade formed by Ruta, Boen- incongruent placement of Haplophyllum in the phytochemical ninghausenia, Thamnosma, Haplophyllum, and Cneoridium (Ruteae and molecular trees (Fig. 2a and c) might be caused by convergence s.s.; Fig. 4). Similar reductionist trends have been previously re- of similar alkaloids in Haplophyllum and Ruta, rather than long- ported in other angiosperm families and explained in terms of pae- branch attraction between Haplophyllum and Cneoridium. domorphic development caused by the elimination of the last The analysis of phytochemical data is hampered not only by initiated organ (Tucker et al., 1993; Hufford, 1996). An alternative problems of homoplasy, but also by methodological complications, interpretation, also consistent with the optimizations of floral mer- for example: (i) a specific compound can be detected only when a ism on the molecular topology (Fig. 4), assumes that the ancestor sufficient amount of it is present in the plant (Price, 1963); and (ii) of Ruteae s.s. was polymorphic for floral merism, thus implying chemotaxonomic reports for Rutaceae rarely mention the names of that Ruta retained the ancestral polymorphisms for all floral whorls

G. Salvo et al. / Molecular Phylogenetics and Evolution xxx (2008) xxx–xxx 11

Sepals 4/5 4 4 5 4 5 5. Conclusion Petals 4/5 4 4 5 4 5 Stamens 8/10 8 8 10 8 10 The finding that traditionally important taxonomic characters R B T H C O do not provide support for the clades identified in molecular phylogenies has become a frequent occurrence with the widespread development of molecular systematics (e.g., Lavin et al., 2001; Moylan et al., 2004; Van der Niet et al., 2005; Norup et al., 2006). When mapped onto a molecular phylogeny, characters originally used to build classifications have been found to be plesiomorphic (Lavin et al., 2001), homoplasious (Moylan et al., 2004; Norup et al., 2006), or simply uninformative for diagnosing clades (Van der Niet et al., 2005). It has been remarked that the frequency of homoplasy in traditional taxonomic characters may reflect the fact that they are usually optimized on molecular topologies, thus establishing an a priori bias for homoplasy in non-molecular datasets (Grant, 2003). In other words, homoplasy may in part derive from inappropriate comparisons between classes of characters at Carpels 4/5 4 2 3/4/5 1 5 different hierarchical levels of organization (Doyle, 1996; Minelli, R B T H C O 1998). This consideration may be especially relevant to the homoplasy observed in the mapping of phytochemical characters onto the Ruteae molecular tree, for different biogenetic pathways can produce the same compound (Hegnauer, 1966; Waterman, 1990). Consequently, the biogenetic pathways leading to these compounds, and not the compounds per se, should be examined and scored as character states. Rather than being viewed in a negative sense, the identification of homoplasy in non-molecular characters should be used as a starting point to study its biological and methodological causes, focusing especially on the developmental pathways underlying phenotypic traits and the conflicting assessments of homology often performed when comparing characters at different levels of organization. Fig. 4. Optimizations of transitions between states for the four characters associated with floral merism: number of sepals, petals, stamens, and carpels. The states of each character, representing the number of units within each whorl, are Acknowledgments symbolized by different patterns in the boxes above of the branches; coexistence of different states for the same character within the same taxon is symbolized by a The authors thank Víctor Suárez Santiago (Haplophyllum from circle. Branches with horizontal lines indicate equivocal reconstructions. Name of Spain), Oyumaa Jamsran (H. dauricum from Mongolia), Louis Zelt- taxa as in Fig. 3. See Supplementary data 2 for further details. ner (H. from Iran and help in the field), Laszlo Csiba, Vincent Savo- lainen and Mark Chase (Rutaceae from the Kew Gardens DNA Bank), Jesús Muñoz, Javier Fuertes Aguilar, Pablo Vargas, and Gonz- and Haplophyllum for the gynoecium only. Fixation of the number alo Nieto Feliner (help and material from the Real Jardín Botánico of organs in the perianth and androecium occurred in the lineages of Madrid), Dennis Hansen (useful contacts), Angel B. Fernández leading to Boenninghausenia/Thamnosma and Cneoridium/Haplo- (Ruta microcarpa leaves), Alfredo Valido Amador (R. oreojasme phyllum, respectively, while carpel number was reduced in the and R. pinnata leaves), Peter Endress (help with the German litera- ancestor of Boenninghausenia and Thamnosma and, independently, ture and useful conversations throughout the research), Rolf Rut- in Cneoridium (Fig. 4). Repeated processes of selection leading to ishauser (help on Rutaceae morphology), Mary Endress (material fixation of character states from a polymorphic ancestor have in- from the Missouri Botanical Gardens and comments on an early deed been proposed as a likely evolutionary explanation for multi- draft), Andre Simoes (material from Meise and help in the lab), ple transitions between character states (Brooks, 1996). Thus the Daniel Heinzmann (help in the lab), Guilhem Mansion (help in results of character mapping indicate that the way forward to the field and during the research), Timo van der Niet, and Chloe understand patterns of homoplasy in the floral morphology of Ru- Galley (help with the analyses), Brigitte Marazzi, Alessia Guggis- teae s.s. may lie in detailed comparisons between the ontogenetic berg, and Barbara Keller (help throughout the research and com- trajectories of floral whorls. ments on an early draft), Elena Benetti and Barbara Seitz Within the context of the phylogenetic relationships sup- (precious help in the library). This study was funded by the Swiss ported by molecular data (Figs. 1 and 2a), Ruta, the type genus National Science Foundation Grant No. 3100A0-105495 to E.C. and for the family, can be diagnosed by using a combination of the Institute of Systematic Botany of the University of Zurich. Addi- plesiomorphic, homoplasious, and autapomorphic morphological tional financial support from the Georges-und-Antoine-Claraz- character states, including: obovate petals and short, thick style Schenkung, the Kommission für Reisestipendien der Schweizeris- (both plesiomorphic states, for they are also present in the chen Akademie der Naturwissenschaften (SANW), and the Equal outgroup); cushion-shaped nectary disk, elongate anthers, and Opportunity Office of the University of Zurich are gratefully uncurved embryo (all homoplasious states present also in acknowledged. Haplophyllum); introrse anthers’ opening (a homoplasious state present also in Haplophyllum and Cneoridium); simple stigma shape and seeds angled dorsally (two homoplasious states pres- Appendix A. Supplementary data ent also in Boenninghausenia and Thamnosma, respectively); and dentate or fimbriate petal margins (an autapomorphy for Ruta; Supplementary data associated with this article can be found, in see Supplementary data 2). the online version, at doi:10.1016/j.ympev.2008.09.004.

12 G. Salvo et al. / Molecular Phylogenetics and Evolution xxx (2008) xxx–xxx

References Gray, A.I., Waterman, P.G., 1978. Coumarins of the Rutaceae. Phytochemistry 17, 845–864. Guggisberg, A., Mansion, G., Kelso, S., Conti, E., 2006. Evolution of biogeographic Akaike, H., 1974. A new look at the statistical model identification. IEEE Trans. Auto. patterns, ploidy levels, and breeding systems in a diploid–polyploid species Control. 19, 716–723. complex of Primula. New Phytol. 171, 617–632. Bacchetta, G., Brullo, S., Giusso del Galdo, G., 2006. Ruta lamarmorae (Rutaceae), a Gut, B.J., 1966. Beiträge zur morphologie des gynoeceums und der blütenachse new species from Sardinia. Edinburgh J. Bot. 63, 153–160. einiger Rutaceen. Bot. Jahrb. 85, 10–247. Baker, R.H., Yu, X.B., DeSalle, R., 1998. Assessing the relative contribution of Hart, M.W., Johnson, S.L., Addison, J.A., Byrne, M., 2004. Strong character molecular and morphological characters in simultaneous analysis trees. Mol. incongruence and character choice in phylogeny of sea stars of the Phylogenet. Evol. 9, 427–436. Asterinidae. Invertebr. Biol. 123, 343–356. Baum, D.A., Small, R.L., Wendel, J.F., 1998. Biogeography and floral evolution of Hegnauer, R., 1966. Comparative phytochemistry of alkaloids. In: Swain, T. (Ed.), baobabs (Adansonia, Bombacaceae) as inferred from multiple data sets. Syst. Comparative Phytochemistry. Academic Press, London and New York, pp. 211– Biol. 47, 181–207. 230. Bärner, J., 1927. Serodiagnostische Verwandschaftforschungen innerhalb der Hendy, M.D., Penny, D., 1989. A framework for the quantitative study of Geraniales, Sapindales, Rhamnales und Malvales. Bibl. Bot., Stuttgart 94, 1–39. evolutionary trees. Syst. Zool. 38, 297–309. Bramwell, D., Bramwell, Z.I., 2001. Wild Flowers of the Canary Islands. Editorial Hillis, D.M., 1987. Molecular versus morphological approaches to systematics. Rueda, Madrid, Spain. Annu. Rev. Ecol. Syst. 18, 23–42. Brooks, D.R., 1996. Explanations of homoplasy at different levels of biological Hillis, D.M., 1995. Approaches for assessing phylogenetic accuracy. Syst. Biol. 44, 3– organization. In: Sanderson, M.J., Hufford, L. (Eds.), Homoplasy: The Recurrence 16. of Similarity in Evolution. Academic Press, San Diego, pp. 3–36. Hufford, L., 1996. Ontogenetic evolution, clade diversification, and homoplasy. In: Bull, J.J., Huelsenbeck, J.P., Cunningham, C.W., Swofford, D.L., Waddell, P.J., 1993. Sanderson, M.J., Hufford, L. (Eds.), Homoplasy: The Recurrence of Similarity in Partitioning and combining data in phylogenetic analysis. Syst. Biol. 42, 384– Evolution. Academic Press, San Diego, pp. 271–301. 397. Hughes, C.E., Lewis, G.P., Yomona, A.D., Reynel, C., 2004. Maraniona. A new Cannatella, D.C., Hillis, D.M., Chippindale, P.T., Weigt, L., Rand, A.S., Ryan, M.J., 1998. dalbergioid legume genus (Leguminosae, Papilionoideae) from Peru. Syst. Bot. Phylogeny of frogs of the Physalaemus pustulosus species group, with an 29, 366–374. examination of data incongruence. Syst. Biol. 47, 311–335. Hulsenbeck, J.P., Ronquist, F., 2001. Mr Bayes: Bayesian inference of phylogeny. Chase, M.W., Morton, C.M., Kallunki, J.A., 1999. Phylogenetic relationships of Bioinformatics 17, 754–755. Rutaceae: a cladistic analysis of the subfamilies using evidence from rbcL and Jussieu, A., 1825. Rutaceae: Haplophyllum. Mém. Mus. Hist. Nat. 12, 464. atpB sequence variation. Am. J. Bot. 86, 1191–1199. Kelchner, S.A., 2000. The evolution of non-coding chloroplast DNA and its Corner, E.J.H., 1976. The Seeds of Dicotyledons. Cambridge University Press, application in plant systematics. Ann. Missouri Bot. Gard. 87, 482–498. Cambridge. Kim, S.-C., Crawford, D.J., Jansen, R.K., 1996. Phylogenetic relationships among the Cunningham, C.W., 1997. Is congruence between data partitions a reliable predictor genera of the subtribe Sonchinae (Asteraceae): evidence from ITS sequences. of phylogenetic accuracy? Empirically testing an iterative procedure for Syst. Bot. 21, 417–432. choosing among phylogenetic methods. Syst. Biol. 46, 464–478. Kishino, H., Hasegawa, M., 1989. Evaluation of the maximum likelihood estimate of Da Silva, M.F., Gottlieb, O.R., Ehrendorfer, F., 1988. Chemosystematics of the the evolutionary tree topologies from DNA sequence data, and the branching Rutaceae: suggestions for a more natural taxonomy and evolutionary order in Hominoidea. J. Mol. Evol. 29, 170–179. interpretation of the family. Plant Syst. Evol. 161, 97–134. Kluge, A.G., 1998. Total evidence or taxonomic congruence. cladistics or consensus Darlu, P., Lecointre, G., 2002. When does the incongruence length difference test classification. Cladistics 14, 151–158. fail? Mol. Biol. Evol. 19, 432–437. Kong, Y.-C., Cheng, K.-F., Ng, K.-M., But, P.P.-H., Li, Q., Yu, S.-X., Chang, H.-T., Cambie, de Queiroz, A., 1993. For consensus (sometimes). Syst. Biol. 42, 368–372. R.C., Kinoshita, T., Kan, W.-S., Waterman, P.G., 1986. A chemotaxonomic division de Queiroz, A., Donoghue, M.J., Kim, J., 1995. Separate versus combined analysis of of Murraya based on the distribution of the alkaloids yuehchukene and phylogenetic evidence. Annu. Rev. Ecol. Syst. 26, 657–681. girinimbine. Biochem. Syst. Ecol. 14, 491–497. Dolphin, K., Belshaw, R., Orme, C.D.L., Quicke, D.L.J., 2000. Noise and incongruence: Kron, K.A., Judd, W.S., Crayn, D.M., 1999. Phylogenetic analyses of Andromedeae interpreting results of the incongruence length difference test. Mol. Phylogenet. (Ericaceae subfam. Vaccinioideae). Am. J. Bot. 86, 1290–1300. Evol. 17, 401–406. Lavin, M., Pennington, R.T., Klitgaard, B.B., Sprent, J.I., de Lima, H.C., Gasson, P.E., Donoghue, M.J., Sanderson, M.J., 1992. The suitability of molecular and 2001. The dalbergioid legumes (Fabaceae): delimitation of a pantropical morphological evidence in reconstructing plant phylogeny. In: Soltis, P., monophyletic clade. Am. J. Bot. 88, 503–533. Soltis, D., Doyle, J. (Eds.), Molecular Systematics of Plants. Chapman and Hall, Lee, M.S.Y., 2001. Uninformative characters and apparent conflict between New York, London, pp. 340–368. molecules and morphology. Mol. Biol. Evol. 18, 676–680. Doyle, J.J., 1996. Homoplasy connections and disconnections: genes and species, Lee, M.S.Y., Hugall, A.F., 2003. Partitioned likelihood support and the evaluation of molecules and morphology. In: Sanderson, M.J., Hufford, L. (Eds.), Homoplasy: data set conflict. Syst. Biol. 52, 15–22. The Recurrence of Similarity in Evolution. Academic Press, San Diego, pp. 37–66. Lewis, P.O., 2001. A likelihood approach to estimating phylogeny from discrete Doyle, J.J., Doyle, J.L., Rauscher, J.T., Brown, A.H.D., 2004. Diploid and polyploid morphological character data. Syst. Biol. 50, 913–925. reticulate evolution throughout the history of the perennial soybeans (Glycine Linnaeus, C., 1735. Systema Naturae, sive regna tria naturae, systematice proposita subgenus Glycine). New Phytol. 161, 121–132. per classes, ordines, genera, et species. Apud Theodorum Haak, Leiden. Endress, P.K., 1990. Patterns of floral construction in ontogeny and phylogeny. Biol. Linnaeus, C., 1753. Species Plantarum, exhibentes plantas rite cognitas, ad genera J. Linnaean Soc. 39, 153–175. relatas, cum differentiis specificis, nominibus trivialibus, synonymis selectis, Endress, P.K., 1999. Symmetry in flowers: diversity and evolution. Int. J. Plant Sci. locis natalibus, secundum systema sexuale digestas. Holmiae, Impensis 160, S3–S23. Laurentii Salvii, Stockholm. Engler, A., 1896. Rutaceae. In: Engler, A., Prantl, K. (Eds.), Nat. Pflanzenfam III, vol. 4. Liscombe, D.K., MacLeod, B.P., Loukanina, N., Nandi, O.I., Facchini, P.J., 2005. Engelmann, Leibzig, pp. 95–201. Evidence for the monophyletic evolution of benzylisoquinoline alkaloid Engler, A., 1931. Rutaceae. In: Engler, A., Prantl, K. (Eds.), Die natürlichen biosynthesis in angiosperms. Phytochemistry 66, 2500–2520. Pflanzenfamilien, 2. Aufl. 19a. Engelmann, Leibzig, pp. 187–359. Lomolino, M.V., Riddle, B.R., Brown, J.H., 2006. Biogeography, third ed. Sinauer Farris, J.S., Kallersjo, M., Kluge, A.G., Bult, C., 1994. Testing significance of Associates, Sunderland, USA. incongruence. Cladistics 10, 315–319. Maddison, P.G., Maddison, D.R., 2000. MacClade4: Analysis of Phylogeny and Felsenstein, J., 1978. Cases in which parsimony or compatibility methods will be Character Evolution. Sinauer Associates, Sunderland, USA. positively misleading. Syst. Zool. 27, 401–410. Marazzi, B., Endress, P.K., De Queiroz, L.P., Conti, E., 2006. Phylogenetic relationships Felsenstein, J., 1985. Confidence limits on phylogenies: an approach using the within Senna (Leguminosae, Cassiinae) based on three chloroplast DNA regions: Bootstrap. Evolution 39, 783–791. patterns in the evolution of floral symmetry and extrafloral nectaries. Am. J. Bot. Fish, F., Waterman, P.G., 1973. The chemosystematics of the Zanthoxylum/Fagara 93, 288–303. complex. Taxon 22, 177–203. Mason-Gamer, R.J., Kellogg, E.A., 1996. Testing for phylogenetic conflict among Fitch, W.M., 1971. Towards defining the course of evolution: minimum change for a molecular data sets in the Triticeae (Gramineae). Syst. Biol. 45, 524–545. specific tree topology. Syst. Zool. 20, 406–416. McKey, D., 1980. Origins of novel alkaloid types: a mechanism for rapid phenotypic Gadek, P.A., Fernando, E.S., Quinn, C.J., Hoot, S.B., Terrazas, T., Sheahan, M.C., Chase, evolution of plant secondary compounds. Am. Nat. 115, 754–759. M.W., 1996. Sapindales: molecular delimitation and infraordinal groups. Am. J. Mester, I., Vicol, E.C., 1971. Contribution to the chemotaxonomy of the Bot. 83, 802–811. Haplophyllum genus. Rev. Roum. Biol. 16, 221–233. Gatesy, J., Baker, R.H., 2005. Hidden likelihood support in genomic data: can forty- Minelli, A., 1998. Molecules, developmental modules, and phenotypes: a five wrongs make a right? Syst. Biol. 54, 483–492. combinatorial approach to homology. Mol. Phylogenet. Evol. 9, 340–347. Goldman, N., Anderson, J.P., Rodrigo, A.G., 2000. Likelihood-based tests of topologies Miyamoto, M., Fitch, W.M., 1995. Testing species phylogenies and phylogenetic in phylogenetics. Syst. Biol. 49, 652–670. methods with congruence. Syst. Biol. 44, 64–76. Graham, S.W., Kohn, J.R., Morton, B.R., Eckenwalder, J.E., Barrett, S., 1998. Moore, J.A., 1936. Floral anatomy and phylogeny in the Rutaceae. New Phytol. 35, Phylogenetic congruence and discordance among one morphological and 318–322. three molecular data sets from Pontederiaceae. Syst. Biol. 47, 545–567. Morton, C.M., Grant, M., Blackmore, S., 2003. Phylogenetic relationships of the Grant, V., 2003. Incongruence between cladistic and taxonomic systems. Am. J. Bot. Aurantioideae inferred from chloroplast DNA sequence data. Am. J. Bot. 90, 90, 1263–1270. 1463–1469.

G. Salvo et al. / Molecular Phylogenetics and Evolution xxx (2008) xxx–xxx 13

Mothes, K., 1981. The problem of chemical convergence in secondary metabolism. evidence from three noncoding plastid DNA regions and morphology. Am. J. Sci. Scientists, 323–326. Bot. 91, 1409–1418. Moylan, E.C., Bennett, J.R., Carine, M.A., Olmstead, R.G., Scotland, R.W., 2004. Simões, A.O., Endress, M., van der Niet, T., Kinoshita, L.S., Conti, E., 2006. Is Phylogenetic relationships among Strobilanthes s.l. (Acanthaceae): evidence Mandevilla (Apocynaceae, Mesechiteae) monophyletic? Evidence from five from ITS nrDNA, trnL–F cpDNA, and morphology. Am. J. Bot. 91, 724–735. plastid DNA loci and morphology. Ann. Missouri Bot. Gard. 93, 565–591. Mueller, R.L., Robert Macey, J., Jaekel, M., Wake, D.B., Boore, J.L., 2004. Morphological Song, W.-H., Li, X.-D., Li, X.-W., Huang, H.-W., Li, J.-Q., 2004. Genetic diversity and homoplasy, life history evolution, and historical biogeography of plethodontid conservation strategy of Psilopeganum sinense, a rare species in the three-gorges salamanders inferred from complete mitochondrial genomes. PNAS 101, reservoir area. Biodiver. Sci. 12, 227–236. 13820–13825. Stace, C.A., 2005. Plant taxonomy and biosystematics—does DNA provide all the Muellner, A.N., Vassiliades, D.D., Renner, S.S., 2007. Placing Biebersteiniaceae, a answers? Taxon 54, 999–1007. herbaceous clade of Sapindales, in a temporal and geographic context.. Plant Stevens, P.F., 2001 onwards. Angiosperm Phylogeny Website. Version 8, June 2007. Syst. Evol. doi:10.1007/s00606-007-0546-x. Available online. Available from: . Nandi, O.I., Chase, M.W., Endress, P.K., 1998. A combined cladistic analysis of Swofford, D.L., 1991. When are phylogeny estimates from molecular and angiosperms using rbcL and non-molecular data sets. Ann. Missouri Bot. Gard. morphological data incongruent? In: Miyamoto, M.M., Cracraft, J. (Eds.), 85, 137–214. Phylogenetic Analysis of DNA Sequences. Oxford University Press, New York, Navarro, F.B., Suarez-Santiago, V.N., Blanca, G., 2004. A new species of Haplophyllum pp. 295–333. A. Juss (Rutaceae) from the Iberian peninsula: evidence from morphological, Swofford, D.L., 2001. PAUPÃ4.0b10: Phylogenetic Analysis Using Parsimony (Ã and karyological and molecular analyses. Ann. Bot. 94, 571–582. other methods). Sinauer, Sunderland, USA. Ng, K.M., But, P.P.-H., Gray, A.I., Hartley, T.G., Kong, Y.-C., Waterman, P.G., 1987. The Sytsma, K.J., 1990. DNA and morphology: inference of plant phylogeny. Trends Ecol. biochemical systematics of Tetradium, Euodia, and Melicope and their Evol. 5, 104–110. significance in the Rutaceae. Biochem. Syst. Ecol. 15, 587–593. Taberlet, P., Gielly, L., Pautou, G., Bouvet, J., 1991. Universal primers for Norup, M.V., Dransfield, J., Chase, M.W., Barfod, A.S., Fernando, E.S., Baker, W.J., amplification of 3 non-coding regions of chloroplast DNA. Plant Mol. Biol. 17, 2006. Homoplasious character combinations and generic delimitation: a case 1105–1109. study from the Indo-Pacific arecoid palms (Arecaceae: Areceae). Am. J. Bot. 93, Tang, F., Ye, Q., Yao, X., Huang, H., 2007. Isolation and characterization of 1065–1080. microsatellite loci in Psilopeganum sinense Hemsl. (Rutaceae), an endangered Novacek, M.J., 1994. Morphological and molecular inroads to phylogeny. In: Grande, herb endemic to Yangtze River valley. Molecular Ecology Notes, published L., Rieppel, O. (Eds.), Interpreting the Hierarchy of Nature: From Systematic online, doi: 10.1111/j.1471-8286.2007.01933.x. Patterns to Evolutionary Process Theories. Academic Press Inc., San Diego, pp. Tilak, V.D., Nene, P.M., 1978. Floral anatomy of the Rutaceae. Ind. J. Bot. 1, 83– 85–131. 90. Patterson, C., 1988. Molecules and Morphology in Evolution: Conflict or Townsend, C.C., 1968. Rutaceae. In: Tutin, T.G. (Ed.), Flora Europea, vol.2, Rosaceae Compromise? Cambridge University Press, Cambridge. to Umbelliferae. Cambridge University Press, Cambridge, pp. 227–230. Porter, J.M., Johnson, L.A., 2000. A phylogenetic classification of Polemoniaceae. Townsend, C.C., 1986. Taxonomic revision of the genus Haplophyllum (Rutaceae). Aliso 19, 55–91. Hooker’s Icones Plantarum, Vol. XL, parts I, II, and III. Bentham-Moxon Trustees, Posada, D., Crandall, K.A., 1998. Modeltest: testing the model of DNA substitution. Kent, UK. Bioinformatics 14, 817–818. Tucker, S.C., Douglas, A.W., Liang, H.-X., 1993. Utility of ontogenetic and Price, J.R., 1963. The distribution of alkaloids in the Rutaceae. In: Swain, T. (Ed.), conventional characters in determining phylogenetic relationships of Chemical Plant Taxonomy. Academic Press, London and New York, pp. 429–452. Saururaceae and Piperaceae (Piperales). Syst. Bot. 18, 614–641. Quicke, D.L.J., Jones, O.R., Epstein, D.R., 2007. Correcting the problem of false Van der Niet, T., Linder, P.H., Bytebier, B., Bellstedt, D.U., 2005. Molecular markers incongruence due to noise imbalance in the incongruence length difference reject monophyly of the subgenera of Satyrium (Orchidaceae). Syst. Bot. 30, (ILD) test. Syst. Biol. 56, 496–503. 263–274. Reimann, A., Nurhayati, N., Backenköhler, A., Oberand, D., 2004. Repeated evolution Wahlberg, N., Nylin, S., 2003. Morphology versus molecules: resolution of the of the pyrrolizidine alkaloid-mediated defense system in separate angiosperm positions of Nymphalis, Polygonia, and related genera (Lepidoptera: lineages. Plant Cell 16, 2772–2784. Nymphalidae). Cladistics 19, 213–223. Rieseberg, L.H., Whitton, J., Linder, C.R., 1996. Molecular marker incongruence in Waterman, P.G., 1975. Alkaloids of the Rutaceae: their distribution and systematic plant hybrid zones and phylogenetic trees. Acta Bot. Neerl. 45, 243–262. significance. Biochem. Syst. Ecol. 3, 149–180. Roalson, E.H., Boggan, J.K., Skog, L.E., Zimmer, E.A., 2005. Untangling Gloxinieae Waterman, P.G., 1983. Phylogenetic implications of the distribution of secondary (Gesneriaceae). I. Phylogenetic patterns and generic boundaries inferred from metabolites within the Rutales. In: Waterman, P.G., Grundon, M.F. (Eds.), nuclear, chloroplast, and morphological cladistic datasets. Taxon 54, 389–410. Chemistry and Chemical Taxonomy of the Rutales. Academic Press, London, pp. Ronquist, F., Huelsenbeck, J.P., 2003. MrBayes 3: Bayesian phylogenetic inference 377–400. under mixed models. Bioinformatics 19, 1572–1574. Waterman, P.G., 1990. Chemosystematics of the Rutaceae: comments on the Rønsted, N., Salvo, G., Savolainen, V., 2007. Biogeographical and phylogenetic interpretation of Da Silva et al. Plant Syst. Evol. 173, 39–48. origins of African fig species (Ficus section Galoglychia). Mol. Phylogenet. Evol. Waterman, P.G., 1998. Alkaloid chemosystematics. In: Cordell, G.A. (Ed.), The 43, 190–201. Alkaloids: Chemistry and Biology. Academic Press, USA, pp. 537–565. Rutschmann, F., Eriksson, T., Abu Salim, K., Conti, E., 2007. Assessing calibration Waterman, P.G., Grundon, M.F., 1983. Chemistry and Chemical Taxonomy of the uncertainty in molecular dating: the assignment of fossils to alternative Rutales. Academic Press, London. calibration points. Syst. Biol. 56, 591–608. Wiens, J.J., Chippindale, P.T., Hillis, D.M., 2003. When are phylogenetic analyses Samuel, R., Ehrendorfer, F., Chase, M.W., Greger, H., 2001. Phylogenetic analyses of misled by convergence? A case study in Texas cave salamanders. Syst. Biol. 52, the Aurantioideae (Rutaceae) based on non-coding plastid DNA sequences and 501–514. phytochemical features. Plant Biol. 3, 77–87. Wiens, J.J., Fetzner, J.W., Parkinson, C.L., Reeder, T.W., 2005. Hylid frog phylogeny San Miguel, E., 2003. Rue (Ruta L., Rutaceae) in traditional Spain: frequency and and sampling strategies for speciose clades. Syst. Biol. 54, 719–748. distribution of its medicinal and symbolic applications. Econ. Bot. 57, 231–244. Wink, M., 2003. Evolution of secondary metabolites from an ecological and Sanderson, M.J., Hufford, L., 1996. Homoplasy: The Recurrence of Similarity in molecular phylogenetic perspective. Phytochemistry 64, 3–19. Evolution. Academic Press, San Diego, USA. Wink, M., Witte, L., 1983. Evidence for a wide spread occurrence of the genes of Sang, T., Crawford, D.J., Stuessy, T.F., 1997. Chloroplast DNA phylogeny, reticulate quinolizidine alkaloid biosynthesis. Induction of alkaloid accumulation in cell evolution, and biogeography of Paeonia (Paeoniaceae). Am. J. Bot. 84, 1120– suspension cultures of alkaloid-‘‘free” species. FEBS Lett. 159, 196–200. 1136. Won, H., Renner, S.S., 2003. Horizontal gene transfer from flowering plants to Saunders, E.R., 1934. On carpel polymorphism. IV. Ann. Bot. 48, 643–692. Gnetum. PNAS 100, 10824–10829. Scholz, H., 1964. Rutales. In: Melchior, H., Wedermann, E. (Eds.), A. Engler’s Syllabus Wortley, A.H., Scotland, R.W., 2006. Determining the potential utility of datasets for der Pflanzenfamilien, vol. 2, twelveth ed. Bornträger, Berlin, pp. 262–277. phylogeny reconstruction. Taxon 55, 431–442. Scott, K.D., McIntyre, C.L., Playford, J., 2000. Molecular analyses suggest a need for a Yoder, A.D., Irwin, J.A., Payseur, B.A., 2001. Failure of the ILD to determine data significant rearrangement of Rutaceae subfamilies and a minor reassessment of combinability for slow loris phylogeny. Syst. Biol. 50, 408–424. species relationships within Flindersia. Plant Syst. Evol. 223, 15–27. Young, N.D., Healy, J., 2003. Gapcoder automates the use of indel characters in Shimodaira, H., Hasegawa, M., 1999. Multiple comparisons of log-likelihoods with phylogenetic analysis. BMC Bioinform. 4, 6. applications to phylogenetic inference. Mol. Biol. Evol. 16, 1114–1116. Zakaria, M.B., 2001. The phytochemistry of Rutaceae species with special reference Simões, A., Endress, M., van der Niet, T., Conti, E., Kinoshita, L.S., 2004. Tribal and to Melicope. Malayan Nat. J. 55, 241–250. intergeneric relationships of Mesechiteae (Apocynoideae, Apocynaceae):

Please cite this article in press as: Salvo, G. et al., Phylogenetic relationships of Ruteae (Rutaceae): New evidence from ..., Mol. Phylogenet. Evol. (2008), doi:10.1016/j.ympev.2008.09.004